Analysis of the Dynamic Modes of Transonic Flow Past a Cylinder
Transonic Flow; Computational Aeroacoustics; Dynamic Modes Decomposition;Proper Orthogonal Decomposition; Flow Past a Cylinder; Inviscid Flow.
The most common approaches for computational aeroacoustics are the direct noise computation through the solution of the governing equations and the use of linear equations to describe sound propagation. An interesting alternative to these methods is a reduced-order model (ROM) approach. The Dynamic Modes Decomposition (DMD) and the Proper Orthogonal Decomposition (POD) are data driven analysis tools that have been used recently as the basis of ROMs. With that in mind, this dissertation aims to use the POD and DMD methods to study the transonic flow past a 2D circular cylinder with Mach numbers 0.5 and 0.75 and modelled with the Euler equations. The flow past the cylinder was chosen as the subject of this work because it is a relatively simple flow, especially for $M_{\infty}=0.5$ but presents rich dynamics. The acceleration of the flow as it passes around the cylinder causes the formation of shock waves, and the resulting adverse pressure gradient causes the separation of the flow. As vortices are emitted and form a von Karman street at the wake, the shock waves also detach and propagate through the domain as sound waves. The Strouhal numbers for both cases were around 0.2, the well-known value for the flow around a cylinder. The POD and DMD are methods that decompose a data set, in this case composed of pressure field snapshots, into coherent structures, or modes. The POD modes are spatially orthogonal structures, while the DMD modes have specific frequencies and rates of growth. The analysis of the modes' structures revealed that the cylinder emits noise in a manner similar to an acoustic dipole, with contributions from other multipoles. Furthermore, the modes for $M_{\infty}=0.75$ revealed that the vortex wake is also a noise source in this case. The DMD decomposition also provided frequencies associated with the modes that matched the peaks in the pressure spectra. The POD and DMD modes were also used to reconstruct the flow fields. The reconstruction with the POD agreed almost exactly with the original data. The results with the DMD, on the other hand, had significant discrepancies. For $M_{\infty}=0.5$ the error for the pressure fluctuations root mean square value was around 15\% to 25\% for most of the domain and had a maximum value of over 200\%. The results showed that the DMD is an analysis tool capable of revealing important information about the system but the POD is a better method to reconstruct the data set.